Types of ML

"Exploring the diverse types of machine learning—supervised, unsupervised, and reinforcement—unlocks a world of possibilities, empowering learners to solve real-world challenges and innovate." 🌟

1. Supervised learning

Gartner, a business consulting firm, predicts that supervised learning will remain the most utilized machine learning among enterprise information technology leaders in 2022. This type of machine learning feeds historical input and output data in machine learning algorithms, with processing between each input/output pair that allows the algorithm to shift the model to create outputs as closely aligned with the desired result. Common algorithms used during supervised learning include neural networks, decision trees, linear regression, and support vector machines.

This machine learning type got its name because the machine is “supervised” while learning, which means you’re feeding the algorithm information to help it learn. The outcome you provide the machine is labelled data, and the rest of the information you give is used as input features. 

For example, suppose you were trying to learn about the relationships between loan defaults and borrower information. In that case, you might provide the machine with 500 cases of customers who defaulted on their loans and another 500 who didn't. The labelled data "supervises" the machine to determine your desired information.

Supervised learning is effective for various business purposes, including sales forecasting, inventory optimisation, and fraud detection. Some examples of use cases include:

• Predicting real estate prices.

• Classifying whether bank transactions are fraudulent or not.

• Finding disease risk factors.

• Determining whether loan applicants are low-risk or high-risk.

• Predicting the failure of industrial equipment's mechanical parts.

2. Unsupervised learning

While supervised learning requires users to help the machine, unsupervised learning doesn't use the same labelled training sets and data. Instead, the machine looks for less obvious patterns in the data. This machine learning type is very helpful when you must identify patterns and use data to make decisions. Common algorithms used in unsupervised learning include Hidden Markov models, k-means, hierarchical clustering, and Gaussian mixture models.

Using the example from supervised learning, let's say you didn't know which customers did or didn't default on loans. Instead, you'd provide the machine with borrower information, and it would look for patterns between borrowers before grouping them into several clusters.

This type of machine learning is widely used to create predictive models. Common applications also include clustering, which creates a model that groups objects based on specific properties, and association, which identifies the rules between the clusters. A few example use cases include:

• Creating customer groups based on purchase behaviour.

• Grouping inventory according to sales and/or manufacturing metrics.

• Pinpointing associations in customer data (for example, customers who buy a specific style of handbag might be interested in a specific style of shoe).

3. Reinforcement learning

Reinforcement learning is the closest machine learning type to how humans learn. The algorithm or agent learns by interacting with its environment and getting a positive or negative reward. Common algorithms include temporal difference, deep adversarial networks, and Q-learning.

Returning to the bank loan customer example, you might use a reinforcement learning algorithm to examine customer information. If the algorithm classifies them as high-risk and they default, the algorithm gets a positive reward. If they don't default, the algorithm gets a negative reward. Ultimately, both instances help the machine learn by better understanding the problem and environment.

Gartner notes that most ML platforms don't have reinforcement learning capabilities because they require higher computing power than most organisations have [2]. Reinforcement learning is applicable in areas capable of being fully simulated that are either stationary or have large volumes of relevant data. Because this type of machine learning requires less management than supervised learning, it’s viewed as easier to work with when dealing with unlabelled data sets. Practical applications for this type of machine learning are still emerging. Some examples of uses include:

• Teaching cars to park themselves and drive autonomously.

• Dynamically controlling traffic lights to reduce traffic jams.

• Training robots to learn policies using raw video images as input that they can use to replicate the actions they see.